Two-stage three-dimensional scene modeling method

Abstract

The invention discloses a two-stage three-dimensional scene modeling method, belonging to the field of computer vision. The method comprises the following the steps that RGB data and depth data are collected at the same time through an RGBD camera and a monocular camera; and modeling is conducted at different speeds through a three-dimensional reconstruction algorithm based on visual synchronous positioning and mapping and a three-dimensional reconstruction algorithm based on a motion recovery structure. In the first stage, an SLAM reconstruction progress is recorded in real time, image frame registration, attitude estimation, point cloud matching, loopback detection and model fusion are carried out, a sparse reconstruction model is generated in a short time, real-time positioning and navigation can be helped, and the time sequence performance of data is enhanced; and in the second stage, the three-dimensional reconstruction algorithm based on the motion recovery structure is carried out through a large number of high-definition RGB images, and a dense and complete reconstruction model with rich details is generated. The method meets the requirement for auxiliary positioning of an reconstruction result in a short time, has timeliness, and can provide a dense reconstruction result with high reconstruction precision.

Description

technical field [0001] The invention belongs to the field of three-dimensional imaging, in particular, to a three-dimensional reconstruction method of an indoor dynamic scene based on an RGBD camera. Background technique [0002] The emergence of Simultaneous Localization and Mapping (SLAM) technology has well solved the problems in the fields of intelligent robot image recognition, mapping and positioning, using sensors mounted on the robot (such as binocular and monocular cameras) to obtain external information, use this information to construct a map, and estimate its own position and attitude. [0003] RGBD cameras can help do better reconstruction, because with sensors that can obtain depth information, you can real-time or save good depth information, and then through some calculations, better 3D reconstruction. [0004] Structure from Motion (SFM) is a 3D reconstruction method that derives 3D information from time-series 2D images. The advantage is that as many pictu...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to propose a two-stage three-dimensional scene reconstruction method for the existing three-dimensional reconstruction method under the problem that the accuracy and time of the scene cannot be guaranteed at the same time, which uses RGBD cameras and optical cameras to perform three-dimensional reconstruction of the indoor environment

Images